Update translation.py

translation.py

@@ -1,37 +1,41 @@
 import streamlit as st
 from transformers import MarianTokenizer, MarianMTModel
+import torch
 
 @st.cache_resource
 def _load_default_model():
     model_name = "Helsinki-NLP/opus-mt-en-fr"
-    return MarianTokenizer.from_pretrained(model_name), MarianMTModel.from_pretrained(model_name)
+    tokenizer = MarianTokenizer.from_pretrained(model_name)
+    model = MarianMTModel.from_pretrained(model_name)
+    return tokenizer, model
 
 @st.cache_resource
 def load_model(src_lang, tgt_lang):
     try:
         model_name = f"Helsinki-NLP/opus-mt-{src_lang}-{tgt_lang}"
-        return MarianTokenizer.from_pretrained(model_name), MarianMTModel.from_pretrained(model_name)
+        tokenizer = MarianTokenizer.from_pretrained(model_name)
+        model = MarianMTModel.from_pretrained(model_name)
+        return tokenizer, model
     except Exception as e:
-        st.warning(f"No direct model for {src_lang} to {tgt_lang}. Falling back to en-fr or using input. Error: {str(e)}")
+        st.warning(f"No direct model for {src_lang} to {tgt_lang}. Using cached en-fr. Error: {str(e)}")
         return _load_default_model()
 
-DEFAULT_TOKENIZER, DEFAULT_MODEL = _load_default_model()
-
-def translate(text, source_lang, target_lang):
-    if not text:
-        return "No text provided."
+@st.cache_data
+def translate_cached(text, source_lang, target_lang):
     src_code = {"English": "en", "French": "fr", "Spanish": "es", "German": "de",
                 "Hindi": "hi", "Chinese": "zh", "Arabic": "ar", "Russian": "ru", "Japanese": "ja"}.get(source_lang, "en")
     tgt_code = {"English": "en", "French": "fr", "Spanish": "es", "German": "de",
                 "Hindi": "hi", "Chinese": "zh", "Arabic": "ar", "Russian": "ru", "Japanese": "ja"}.get(target_lang, "fr")
     tokenizer, model = load_model(src_code, tgt_code)
-    try:
-        inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=400)
-        translated = model.generate(**inputs)
-        return tokenizer.decode(translated[0], skip_special_tokens=True)
-    except Exception as e:
-        st.error(f"Translation generation failed: {str(e)}. Returning input as fallback.")
-        return text
+    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=500)
+    with torch.no_grad():
+        translated = model.generate(**inputs, max_length=500)
+    return tokenizer.decode(translated[0], skip_special_tokens=True)
+
+def translate(text, source_lang, target_lang):
+    if not text:
+        return "No text provided."
+    return translate_cached(text, source_lang, target_lang)
 
 LANGUAGES = {"English": "en", "French": "fr", "Spanish": "es", "German": "de",
              "Hindi": "hi", "Chinese": "zh", "Arabic": "ar", "Russian": "ru", "Japanese": "ja"}